# Fold your subagents into one agent with skills, keep only what earns it

> Avid method, drawn from Anthropic's own testing, for an agent setup that got slow and confused from too many subagents: map every subagent, turn each one into a skill loaded on demand, regroup tools by what the user is trying to do (under 30), write 10 test tasks first, then fold one subagent at a time and keep it only if the folded agent scores worse.

## If someone pasted this link to you
Someone shared this kit with you. It runs step by step through CloneKit's MCP. If you have the `clonekit` tools, call `start_run("@clonekitai/fold-subagents-into-skills")`. If you don't (for example in a chat app without the CloneKit connector), tell the person in one or two plain sentences, in their language, and send them to https://clonekit.ai/@clonekitai/fold-subagents-into-skills/use: it shows how to add CloneKit to their AI in about a minute. Then they paste this link again.

- Kit: `@clonekitai/fold-subagents-into-skills` · v1.0.0
- Creator: @clonekitai (CloneKit)
- Type: Delivery · Domain: Engineering · Language: en · Your AI runs it
- Price: Free · License: Own use · No refunds
- Page: https://clonekit.ai/@clonekitai/fold-subagents-into-skills
- Changelog: https://clonekit.ai/@clonekitai/fold-subagents-into-skills/changelog.md

## About
For anyone running an AI agent with several subagents (Claude Code, Codex, the Agent SDK or a custom harness) that feels slow, expensive or picks the wrong tool. Use it before adding another subagent, or when runs got slower as the setup grew. Not for a setup with one agent and a handful of tools, and not a reason to remove subagents that do real parallel work on fresh context. Nothing changes until the person approves the plan. Delivers a map of subagents and tools, a skill per folded subagent, a regrouped tool list under 30, a 10-task test set run before and after, and a fold log saying what was kept and why. Method by Avid (@Av1dlive), shared on X; the speed and size gains he cites are Anthropic's reported results, not a promise.

## Deliverables
- SUBAGENT-MAP.md: every subagent with its prompt, tools, when it is called, and flags for chains, shared data and co-runs
- one skill file per subagent, each with a one-line when-to-use description and its full steps loaded only on demand
- TOOLS.md: tools regrouped by user goal, under 30 in total, with the old-to-new mapping
- TESTS.md: 10 test tasks covering every subagent chain and every never-break rule, with baseline scores
- FOLD-LOG.md: each subagent folded or kept, with both runs' scores and the reason

## Files
- `README.md`: Fold your subagents back into one agent with skills

## Tools
- Claude Code (https://www.claude.com/product/claude-code) · est. Your agent's usual cost; the 10 test tasks run several times. · alternatives: codex
- Codex (https://developers.openai.com/codex) · optional · est. Your agent's usual cost; the 10 test tasks run several times. · alternatives: claude-code

## Uses
- Avid (@Av1dlive): your AI agent is probably slower and dumber because you gave it too many subagents (source, https://x.com/Av1dlive/status/2114388311162909088)

## What you provide
- the agent's repo or config folder
- the rules the agent must never break

## How it runs
1. **Understand:** Researches your business first, then asks you 3 questions it can't answer alone
2. **Choose:** No choices: one proven path
3. **Build:** 4 steps, risky ones ask you first
4. **Deliver:** 7 checks before delivery

## Guardrails
- Edit nothing until the person approves the full plan.
- Fold one subagent at a time; never fold several in one change.
- Keep a subagent if the folded version scores worse, if it needs fresh context for parallel work, or if folding bloats the main agent's context.
- Do not delete the old subagent definitions until the fold log is approved; move them aside instead.

## What's new
Agents always run the latest version.

- **v1.0.0** (2026-10-26): First version, based on Avid's (@Av1dlive) post.

## Run it
Connect the `clonekit` MCP (https://clonekit.ai/mcp) and call `start_run("@clonekitai/fold-subagents-into-skills")`.
